Season analytics

What the model actually did, graded after the fact. Pregame snapshots are never rewritten.

Model
Graded calls
53
Season to date
Hit rate
55.0%
Recommended sides only
Units
-2.30u
Flat 1u staking
ROI
-20.0%
Return on amount staked
MAE
14.50
Mean absolute projection error
Avg CLV
+0.30
Line value vs close

Weekly performance

Hit rate and units by week.

The dashed line is 52.4% — roughly break-even at standard -110 pricing.

Projection error

Mean absolute error by week. Lower is better.

Confidence buckets

High confidence should outperform low. If it doesn't, confidence is broken.

BucketnHitUnitsROI
high9251%+3.20u150%
medium2958%+1.90u-80%
low8554%-0.50u-120%

Prop type performance

Where the model earns and where it should stay out.

MarketnHitMAEROI
Receptions7362%13.8060%
Anytime TD9261%15.70-50%
Receiving Yards5958%13.4020%
Targets8756%15.20-100%
Passing Yards2554%17.0080%
Pass Attempts7251%12.70-50%
Rushing Yards7645%14.1090%

Calibration

A 60% call should win about 60% of the time.

Signals working / not working

Average edge captured per signal, graded after the fact.

SignalnHitAvg edgeVerdict
weather9160%+5.10Working
expected usage6655%+4.60Working
recent form2554%+4.50Working
market anchor4756%+2.70Working
opponent matchup3948%+1.90Not working

Biggest misses

Largest absolute projection errors. Worth reading before trusting a similar call.

Loading graded projections

These results come from the labelled demo dataset. They demonstrate the grading pipeline, not a real betting record.